JMIR Publications

ABSTRACT

Background:

Finding reliable health information is a major challenge for families of individuals with neurodevelopmental disabilities/differences (NDD). NDD encompasses various conditions affecting up to 14% of children in developed countries and most individuals present with complex phenotypes and related conditions, it is challenging for families to develop literacy solely by searching online information. While in-person coaching has been shown to enhance care, it is still available to a minority of individuals with NDD. Chatbots, or computer programs recapitulating conversation, have emerged in the commercial sector as useful in answering questions but their use in healthcare remains limited.

Objective:

We aimed at identifying the features required for chatbot dealing with complex health-related domains such as NDD, and then develop a chatbot able to provide health-promoting information.

Methods:

We developed, in collaboration with individuals with lived experience, a chatbot able to provide information about trusted resources covering core knowledge and services which may be of interest to families. We used Python's Django framework for the development. We also used a knowledge graph depicting the key entities in NDD and their relations, to allow the chatbot to suggest web resources which may be related to the user queries. To identify NDD domain-specific entities from user input, we used a combination of standard sources (UMLS) and other entities identified by health professionals as well as collaborators. We also developed a user interface and assessed its functionality with experienced individuals.

Results:

We developed a chatbot named CAMI (Coaching assistant for medical/health information) which allowed lay people to enter questions related to NDD.Although most entities were identified in the text, some were not captured in our system and therefore went undetected. Nonetheless, the chatbot could provide resources addressing most user queries related to NDD. We observed that enriching the vocabulary (with synonyms or lay language) for specific subdomains enhanced entity detection in user input. By utilizing a dataset of numerous individuals with NDD, we developed a knowledge graph that established meaningful connections between entities. This allowed the chatbot to present related symptoms, diagnoses, and resources for those conditions.

Conclusions:

To our knowledge, CAMI is the first chatbot providing resources related to NDD. Our work highlighted the importance of engaging end users to supplement standard generic ontologies to named entities for language recognition. We also found that complex medical and other health-related information could be integrated by using knowledge graphs and leveraging existing large datasets. This has multiple implications: generalizability to other health domains but also reducing the need for experts and optimizing their input while keeping healthcare professionals in the loop. Finally, our work also shows how health and computer science domains need to collaborate to achieve the granularity needed to make chatbots truly useful and impactful.